gh-119613: use C99+ functions instead of Py_IS_NAN/INFINITY/FINITE

[skirpichev]

• Issue: Deprecate Py_IS_NAN/INFINITY/FINITE? #119613

[Eclips4]

This partially address #119613. I think this part doesn't require a news entry. CC @Eclips4

Yes, since this is a non-visible user change, I agreed.

[rhettinger]

@tim-one is there an on-going reason to have these portability macros? Are NaN values guaranteed to be supported?

[skirpichev]

Since 194a952 these macros are just aliases for isnan, isinf, etc:

cpython/Include/pymath.h

Lines 30 to 40 in 3e8b609

	// Py_IS_NAN(X)
	// Return 1 if float or double arg is a NaN, else 0.
	#define Py_IS_NAN(X) isnan(X)
	// Py_IS_INFINITY(X)
	// Return 1 if float or double arg is an infinity, else 0.
	#define Py_IS_INFINITY(X) isinf(X)
	// Py_IS_FINITE(X)
	// Return 1 if float or double arg is neither infinite nor NAN, else 0.
	#define Py_IS_FINITE(X) isfinite(X)

[vstinner]

LGTM. Simple text replacement.

[vstinner]

@rhettinger:

@tim-one is there an on-going reason to have these portability macros? Are NaN values guaranteed to be supported?

Python 3.10 and older supported platforms without IEEE 754, without NaN, and without C99 <math.h>. We had complex macros such as:

#ifndef Py_IS_NAN
#if defined HAVE_DECL_ISNAN && HAVE_DECL_ISNAN == 1
#define Py_IS_NAN(X) isnan(X)
#else
#define Py_IS_NAN(X) ((X) != (X))
#endif
#endif

Python 3.11 and newer requires IEEE 754, NaN and C99 <math.h>. Macros were kept for backward compatibility only:

#define Py_IS_NAN(X) isnan(X)

Issues:

• Require IEEE 754 floating point to build Python 3.11 #91073
• Remove the Py_NO_NAN macro: require NAN to build Python 3.11 #90814
• [C API] Py_IS_INFINITY() macro doesn't work in the limited C API if isinf() is not defined #89603
• Python can now use the C99 NAN constant or __builtin_nan() #90798

See also this LWN article: CPython, C standards, and IEEE 754.

[vstinner]

@mdickinson: Are you ok with this change? You wrote a comment about these macros there: #119457 (comment)

[mdickinson]

@mdickinson: Are you ok with this change?

Yes, absolutely. I'm fairly sure the macros were born out of (now somewhat ancient) portability needs that no longer apply, now that C99 (and later) adoption is sufficiently widespread.

I'd suggest running all buildbots on this PR, just to be on the safe side.

[mdickinson]

LGTM

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @mdickinson for commit 13dcd12 🤖

If you want to schedule another build, you need to add the 🔨 test-with-buildbots label again.

[skirpichev]

There are tests failures, but hardly it's related to the pr. Something expected, given there are not build bots for non-IEEE-754 platforms, right?

[mdickinson]

There are tests failures, but hardly it's related to the pr.

Agreed - the failures look unrelated.

[vstinner]

Merged, thanks.